Improved Comprehensibility and Reliability of Explanations via Restricted Halfspace Discretization